Out-licensing

Accelerated Tomographic Image Reconstruction Algorithms on Graphics Processing Units (GPU)

Researchers in the Molecular Imaging Instrumentation Laboratory at Stanford University have developed an efficient new system for tomographic image reconstruction that is 10 - 50 times less expensive than current CPU-based computational methods with equivalent computational power. The technology utilizes Graphics Processing Unit (GPU) hardware and software libraries to accelerate the line back projection and forward projection steps of tomographic image reconstruction. The 3D OSEM (ordered-subset expectation-maximization) algorithm was implemented on the GPU and tested with data acquired a commercially available positron emission tomography (PET) scanner. The GPU results demonstrated an acceleration factor of 30 compared to an implementation of 3D-OSEM on a single CPU. The invention could also be applied to single photon emission computed tomography (SPECT), X-ray CT and other imaging modalities that rely on line projection steps for tomographic image reconstruction.

In addition to optimizing certain aspects of the code to improve computational efficiency, the inventors are currently working on incorporating accurate modelling of various image-degrading effects into the image reconstruction process. This will improve reconstructed image quality and quantification further by correcting for these effects.

Conclusion

We have developed accelerated tomographic image reconstruction algorithms on GPU for medical imaging.

Relevance/Opportunity

Please enquire quoting reference no. S05-411 regarding licensing or codevelopment partnerships.

Preclinical